SS-DPPN: A self-supervised dual-path foundation model for the generalizable cardiac audio representation